def load_embeddings()

in empchat/models.py [0:0]

• 27 lines of code
• 5 McCabe index (conditional complexity)

def load_embeddings(opt, dictionary, model):
    path = opt.embeddings
    logging.info(f"Loading embeddings file from {path}")
    emb_table = model.embeddings.weight
    requires_grad = emb_table.requires_grad
    emb_table[dictionary[PAD_TOKEN]].zero_()  # Zero-out padding index
    n_added = 0
    missing_dict = set(dictionary.keys())
    with open(path) as f:
        for line in f:
            parsed = line.rstrip().split(" ")
            assert len(parsed) == opt.embeddings_size + 1
            w = parsed[0]
            if w in dictionary:
                n_added += 1
                vec = torch.Tensor([float(i) for i in parsed[1:]])
                if opt.normalize_emb:
                    vec = vec / vec.norm(2)
                emb_table.data[dictionary[w]] = vec
                missing_dict.remove(w)
    sample = ", ".join(list(missing_dict)[:8])
    logging.info(
        f"Loaded {n_added} vectors from embeddings file; {len(missing_dict)} are "
        f"missing, among which: {sample}"
    )
    emb_table.detach_()
    emb_table.requires_grad = requires_grad